Bug 132047

Summary: Sunspider 1.0.2 - date-format-tofte getting bad results in chrome, compared with other browsers
Product: WebKit Reporter: beni <beni.duca>
Component: Tools / TestsAssignee: Nobody <webkit-unassigned>
Status: RESOLVED INVALID    
Severity: Normal CC: barraclough, commit-queue, ggaren, mjs, ossy
Priority: P2    
Version: 528+ (Nightly build)   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
Fix to date-format-tofte test none

Description beni 2014-04-23 01:10:25 PDT 
Created attachment 229962 [details]
Fix to date-format-tofte test

It seems that chrome 34 (also previous versions) is getting really bad results compared with other browsers. Here is a benchmark result : http://screencast.com/t/ljWBY5x6i

I investigated the problem and seems that "date-format-tofte" is using the "eval" method in its code to call some functions. I think this is not a test for evaluating the "eval" method(Anyway is a bad habit and not recommended to use it. ), so it should be approach differently. I attached a slightly different  version of the test, which fixed that issue. Hope it helps.
Comment 1 WebKit Commit Bot 2014-04-23 01:13:18 PDT 
Attachment 229962 [details] did not pass style-queue:


Total errors found: 0 in 0 files


If any of these errors are false positives, please file a bug against check-webkit-style.
Comment 2 Csaba Osztrogonác 2014-04-23 01:54:22 PDT 
I got the following results on my desktop computer (Windows, Core i5)

Google Chrome - 34.0.1847.116 m:
format-tofte:        9.7ms +/- 5.0%

Firefox 28.0:
format-tofte:       11.4ms +/- 6.1%


It seems there is a bug in your Chrome/V8 version. But I don't think if
we should modify the SunSpider benchmark to fix a bug in V8. I suggest
you should report this bug to V8 bug tracker instead.
Comment 3 Alexey Proskuryakov 2014-04-24 09:54:12 PDT 
> I don't think if we should modify the SunSpider benchmark to fix a bug in V8.

Yes, Sunspider is meant to test real life code as it exists on the web. It's not meant to be hand-tuned to artificially improve V8 results.
Comment 4 beni 2014-04-25 00:11:03 PDT 
It's OK, I found out that AVAST is the reason i got those results, even i was disable it during the test (also I have done the test on multiple machines - Windows - but apparently all had AVAST), so after uninstalling and restart, the tests are OK. So sorry if i took you too much time.

Anyway, using eval method it's not professional and not recommended and hard to debug, especially in a so easy to avoid case, but after all is a test.

Also congrats for the great work you do there :).

Thanks.